PLDPNet: End-to-end hybrid deep learning framework for potato leaf disease prediction

Abstract

Agricultural productivity plays a vital role in global economic development and growth. When crops are affected by diseases, it adversely impacts a nation’s economic resources and agricultural output. Early detection of crop diseases can minimize losses for farmers and enhance production. In this study, we propose a new hybrid deep learning model, PLDPNet, designed to automatically predict potato leaf diseases. The PLDPNet framework encompasses image collection, pre-processing, segmentation, feature extraction and fusion, and classification. We employ an ensemble approach by combining deep features from two well-established models (VGG19 and Inception-V3) to generate more powerful features. The hybrid approach leverages the concept of vision transformers for final prediction. To train and evaluate PLDPNet, we utilize the public potato leaf dataset: early blight, late blight, and healthy leaves. Utilizing the strength of segmentation and fusion feature, the proposed approach achieves an overall accuracy of 98.66%, and F1-score of 96.33%. A comprehensive validation study is conducted using Apple (4 classes) and tomato (10 classes) datasets achieving impressive accuracies of 96.42% and 94.25%, respectively. These experimental findings confirm that the proposed hybrid framework provides more effective and accurate detection and prediction of potato crop diseases, making it a promising candidate for practical applications.

Keywords

• Potato leaf disease
• Plant leaf disease prediction
• Hybrid prediction AI mode
• Deep learning
• Feature fusing and concatenation
• Vision Transformer (ViT)